In certain OpenGL workloads, the performance of OpenSWR can even beat out that of LLVMpipe. Back in October was when Intel originally published the OpenSWR bits for Mesa, which we covered in an article entitled Intel Is Making A High-Performance Software Rasterizer For Mesa.
Since that original patch series, the developers have been working on better conformance, more features, supporting Mesa's latest code-base, and supporting LLVM 3.6~3.8.
Intel's Tim Rowley described OpenSWR for Mesa, for those that didn't read about it the first time:
This patch series adds the OpenSWR driver, a new software rasterizer project by Intel. The goal of this project is to create a high performance, highly scalable renderer targeted towards visualization workloads. For geometry heavy workloads we see a considerable speedup over llvmpipe, which is to be expected as the geometry frontend of llvmpipe is single threaded.
Our rasterizer is x86 specific and requires AVX or AVX2. The driver fits into the gallium framework, and reuses gallivm for doing the TGSI to vectorized llvm-IR conversion of the kernel of the shaders.
Adding OpenSWR to Mesa adds over 54,000 lines of new code. The current patch series can be found here but so far there is no other developer upstream feedback on the latest work.